yes

The reviewer's argument doesnt make any sense, since a mediation analysis is a regression and a regression assess associations between variables. We are always looking for associations.

But a problem of cross sectional analyses is that you cannot derive causal relationships from the statistical model alone. There are many equivalent models, which all may be significant. Therefore, causality, which is implied by many researchers when conducting mediation analyses, has to be established by strong theory in the first place.

For example, is plausible to hypothesize that a personality trait (which is quite stable) may influence causally reaction times in an experiment. You could quantify this association by a correlation (or total effect).

But it is quite implausible to assume that the same association has a reversed cause, i.e. that reaction times influence the personality trait, especially, if this was measures before the experiment.

Therefore, you should have a strong theory AND you should directly show the implied associations between ALL possible variables, measured AND unmeasured. You should use directed acyclic graphs (DAG) to do it. The advantage is that DAGs also show for which variables you COULD (covariables), SHOULD (confounders) or MUST NOT (colliders, collider bias) adjust, if you want an unbiased estimate of your effect (total, direct or indirect).

i think longitudinal studies may have possbly stronger evidence, but it is not true that you cannot make causal inferences from cross sectional studies. Please have a look at publications by

Judea Pearl (The Book of Why, Causality, Causal Inference in Statistics)

Imbens & Rubin (Causal Inference)

Scott Cunningham (Causal Infernce The Mixtape)